POC Prison: Why agentic systems never escape the lab and how to fix that in 90 days - Luise Freese
This title could be clearer and more informative.Try out Clickbait Shieldfor free (5 uses left this month).
A conference talk transcript addressing why agentic AI systems get stuck in proof-of-concept (POC) stages and never reach production. The speaker argues the root causes are organizational and cultural rather than technical: unclear ownership, fragmented budgets, immature governance, poor data quality, and a culture that rewards demos over deployments. A 90-day structured program is proposed to escape 'POC prison': the first 30 days document real workflows and blockers, the next 30 days establish accountability and a path to production for one initiative, and the final 30 days deploy into real systems with real users and live monitoring. Key principles include governance-as-code embedded in delivery pipelines, bounded agents with explicit authority, named business owners, and treating compliance as an accelerator rather than a veto. The talk emphasizes that lowering the barrier to write code with AI doesn't eliminate the need for solid software engineering practices.